PostgreSQL: ERROR: 42601: a column definition list is required for functions returning "record"
Solution 1:
Return selected columns
CREATE OR REPLACE FUNCTION get_user_by_username(_username text
, _online bool DEFAULT false)
RETURNS TABLE (
user_id int
, user_name varchar
, last_activity timestamptz
)
LANGUAGE plpgsql AS
$func$
BEGIN
IF _online THEN
RETURN QUERY
UPDATE users u
SET last_activity = current_timestamp -- ts with time zone
WHERE u.user_name = _username
RETURNING u.user_id
, u.user_name
, u.last_activity;
ELSE
RETURN QUERY
SELECT u.user_id
, u.user_name
, u.last_activity
FROM users u
WHERE u.user_name = _username;
END IF;
END
$func$;
Call:
SELECT * FROM get_user_by_username('myuser', true);
You had DECLARE result record;
but didn't use the variable. I deleted the cruft.
You can return the record directly from the UPDATE
, which is much faster than calling an additional SELECT
statement. Use RETURN QUERY
and UPDATE
with a RETURNING
clause.
If the user is not _online
, default to a plain SELECT
. This is also the (safe) default if the second parameter is omitted - which is only possible after providing that default with DEFAULT false
in the function definition.
If you don't table-qualify column names (tablename.columnname
) in queries inside the function, be wary of naming conflicts between column names and named parameters, which are visible (most) everywhere inside a function.
You can also avoid such conflicts by using positional references ($n
) for parameters. Or use a prefix that you never use for column names: like an underscore (_username
).
If users.username
is defined unique in your table, then LIMIT 1
in the second query is just cruft. If it is not, then the UPDATE
can update multiple rows, which is most likely wrong. I assume a unique username
and trim the noise.
Define the return type of the function (like @ertx demonstrated) or you have to provide a column definition list with every function call, which is awkward.
Creating a type for that purpose (like @ertx proposed) is a valid approach, but probably overkill for a single function. That was the way to go in old versions of Postgres before we had RETURNS TABLE
for that purpose - like demonstrated above.
You do not need a loop for this simple function.
Every function needs a language declaration. LANGUAGE plpgsql
in this case.
I use timestamptz
(timestamp with time zone
) instead of timestamp
(timestamp without time zone
), which is the sane default. See:
- Ignoring time zones altogether in Rails and PostgreSQL
Return (set of) whole row(s)
To return all columns of the existing table users
, there is a simpler way. Postgres automatically defines a composite type of the same name for every table. Just use RETURNS SETOF users
to vastly simplify the query:
CREATE OR REPLACE FUNCTION get_user_by_username(_username text
, _online bool DEFAULT false)
RETURNS SETOF users
LANGUAGE plpgsql AS
$func$
BEGIN
IF _online THEN
RETURN QUERY
UPDATE users u
SET last_activity = current_timestamp
WHERE u.user_name = _username
RETURNING u.*;
ELSE
RETURN QUERY
SELECT *
FROM users u
WHERE u.user_name = _username;
END IF;
END
$func$;
Return whole row plus custom addition
To address the question added by TheRealChx101 in a comment below:
What if you also have a calculated value in addition to a whole table? 😑
Not as simple, but doable. We can send the whole row type as one field, and add more:
CREATE OR REPLACE FUNCTION get_user_by_username3(_username text
, _online bool DEFAULT false)
RETURNS TABLE (
users_row users
, custom_addition text
)
LANGUAGE plpgsql AS
$func$
BEGIN
IF _online THEN
RETURN QUERY
UPDATE users u
SET last_activity = current_timestamp -- ts with time zone
WHERE u.user_name = _username
RETURNING u -- whole row
, u.user_name || u.user_id;
ELSE
RETURN QUERY
SELECT u, u.user_name || u.user_id
FROM users u
WHERE u.user_name = _username;
END IF;
END
$func$;
The "magic" is in the function call, where we (optionally) decompose the row type:
SELECT (users_row).*, custom_addition FROM get_user_by_username('foo', true);
db<>fiddle here (showing all)
If you need something more "dynamic", consider:
- Refactor a PL/pgSQL function to return the output of various SELECT queries
Solution 2:
if you would like to create function returning setof record, you'll need to define column types in your select statement
More info
Your query should look something like this:
select * from get_user_by_username('Username', True) as
f(user_id integer, user_name varchar, last_activity, varchar, created date, email archar, approved boolean, last_lockout timestamp, last_login timestamp,
last_password_changed timestamp, password_question varchar, comment varchar)
(you will probably need to change the data types)
I personaly prefer the types approach. it assures that if the function is edited, all the queries will return correct results. It might be a pain because every time you modify function's arguments you'll need to recreate/drop types aswell tho.
Eg:
CREATE TYPE return_type as
(user_id integer,
user_name varchar,
last_activity varchar,
created timestamp,
email varchar,
approved boolean,
last_lockout timestamp ,
last_login timestamp,
last_password_changed timestamp,
password_question varchar,
comment varchar);
create or replace function get_user_by_username( username varchar(250), online
boolean) returns setof return_type as $$
declare _rec return_type;
begin
if online then
update users
set last_activity = current_timestamp
where user_name = username;
end if;
for _rec in select
user_id,
user_name,
last_activity,
created,
email,
approved,
last_lockout,
last_login,
last_password_changed,
password_question,
comment
from
users
where
user_name = username
limit 1
loop
return next _rec;
end loop
end;
$$ language plpgsql;